LNG anti-rolling prediction method and device based on GA-BP neural network algorithm

Abstract

The invention discloses an LNG anti-tumbling prediction method and device based on a GA-BP neural network algorithm, and the method comprises the steps: carrying out the simulation based on an LNG storage tank two-dimensional model, obtaining the size, diameter, initial density difference and critical density difference corresponding to the layering height of different storage tanks, and constructing a tumbling process data set; establishing an LNG storage tank rolling prediction model based on a GA-BP neural network algorithm, and training the LNG storage tank rolling prediction model by adopting the training set; the optimized and trained LNG storage tank rolling prediction model is adopted for rolling prediction, and the critical density difference is calculated according to the input values of the storage tank volume, diameter, initial density difference and layering height so as to judge whether the LNG storage tank is about to roll or not. The method can be used for predicting the critical density difference of the LNG storage tank so as to ensure safe operation of the LNG storage tank.

Description

Technical Field

[0001] The present invention relates to the field of liquefied natural gas (LNG) storage and transportation, and in particular to a LNG anti-rollover prediction method and device based on a GA-BP neural network algorithm. Background Art

[0002] Natural gas is mainly produced, stored, traded and transported in the form of liquefied natural gas. LNG is an important link in the natural gas industry chain. With the rapid development of the domestic LNG industry, safety issues in its production, storage, transportation and other links have gradually attracted widespread attention. Especially in terms of storage, on the one hand, due to the different sources, components and densities of LNG, tumbling is prone to occur when the storage tank is not fully mixed; on the other hand, because nitrogen has a relatively low boiling point, it will evaporate first under the influence of heat leakage, resulting in density differences and stratified tumbling.

[0003] The study of LNG tank rollover prediction methods is an important basic work to protect the safe operation of LNG receiving stations. In existing research, LNG rollover models are generally established based on physical principles and machine learning technologies. Based on the principles of thermodynamics and the conservation of mass and energy, a lumped parameter model for LNG rollover prediction is established to describe the change law of parameters such as density, temperature and composition during the storage process of LNG. The model is solved by numerical simulation methods to predict the occurrence conditions and evolution process of the rollover phenomenon. On this basis, machine learning algorithms such as neural networks and support vector machines are used to train and learn a large amount of data, extract feature information from the data, and establish an LNG rollover prediction model to predict new data through the model. Among them, the application of back propagation neural networks is the most mature. For example, the invention with the publication number CN118935247A discloses an LNG tank rollover prevention adaptive warning method, which collects static parameters such as the inner diameter, height, thermal conductivity and thickness of cold insulation materials at different positions during the construction of the tank, and processes and displays the spatial distribution of dynamic parameters collected by LTD. The calculation model parameters are calibrated through parameters such as tank area, tank volume, filled volume, heat leakage, static evaporation rate, etc. to ensure the accuracy of static and dynamic parameters. On this basis, the tank prediction model is connected to carry out orthogonal tests on the rollover arrival time and the rollover peak evaporation rate to ensure that the results obtained by the prediction model follow standards, specifications and objective laws. Finally, the prediction results of the traditional LTD warning and anti-rollover prediction model are combined to achieve adaptive warning of tank anti-rollover. However, prediction methods such as BP neural network have disadvantages such as local minima, long training time and slow convergence speed, which affect the prediction results. Summary of the invention

[0004] The purpose of the present invention is to propose a LNG anti-rollover prediction method and device based on the GA-BP neural network algorithm. Aiming at the safety problems caused by the rollover of LNG storage tanks, a LNG storage tank rollover prediction model is established by combining genetic and neural network algorithms. The model is trained using a data set of Fluent numerical simulation of LNG storage tanks. The test results have a small error and can be used to predict the critical density difference of LNG storage tanks to ensure the safe operation of LNG storage tanks.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0006] In a first aspect, the present invention discloses a LNG anti-rollover prediction method based on a GA-BP neural network algorithm, the method comprising the following steps:

[0007] S1, Fluent software is used to construct a two-dimensional model of the LNG storage tank, in which the liquid layers have the same height and the liquid density within the layers is uniform; the two-dimensional model of the LNG storage tank is meshed, and initial conditions and boundary conditions are set;

[0008] S2, based on the simulation of the two-dimensional model of LNG storage tanks, obtain the critical density difference corresponding to different tank volumes, diameters, initial density differences, and layer heights, construct a tumbling process data set, and divide the tumbling process data set into a training set and a test set;

[0009] S3, based on the GA-BP neural network algorithm, the LNG tank rollover prediction model is established, and the training set is used to train the LNG tank rollover prediction model. During the training process, the minimum error between the test value and the expected value of the test sample is used as the objective function to adjust and optimize the parameters of the LNG tank rollover prediction model;

[0010] S4, using the optimized trained LNG tank rollover prediction model to perform rollover prediction, calculate the critical density difference according to the input values ​​of the tank volume, diameter, and layer height to determine whether the LNG tank is about to roll over.

[0011] Step S1 further comprises:

[0012] Intercept two adjacent and equal-height stratified liquids in the LNG storage tank, where the density of the upper LNG layer is less than that of the lower LNG layer, the two liquid strata have the same height, and the density of the liquids in the strata is uniform. Simplify the LNG storage tank into a two-dimensional solid model, and set multiple monitoring points on the two-dimensional solid model.

[0013] Based on the simplified two-dimensional solid model, the Fluent software was used to construct a two-dimensional model of the LNG storage tank based on the mixture model and perform meshing;

[0014] Set the initial conditions and boundary conditions; the initial conditions include the tank volume, diameter, layer height and initial density difference; the boundary condition is set to the wall, and the turbulence equation uses the k-ε equation.

[0015] Step S2 further comprises:

[0016] S21, set the tank volume, diameter and layer height;

[0017] S22, setting multiple groups of initial density differences to simulate the tumbling time and tumbling coefficient of different initial density differences;

[0018] S23, analyzing the variation range of the rollover coefficient corresponding to different initial density differences, and setting the initial density difference when the variation range of the rollover coefficient exceeds a preset variation threshold as the critical density difference for safe storage of the storage tank;

[0019] S24, changing the tank volume, diameter and layer height, repeating steps S22 to S23, until critical density differences corresponding to multiple groups of tank volumes, diameters and layer heights are obtained, and a tumbling process data set is constructed.

[0020] Step S3 further comprises:

[0021] S31, determine the topological structure of the GA-BP neural network;

[0022] S32, encode the original data and establish the initial population; the length N of a single chromosome is:

[0023] N=(n1×n2+n2×n3+n2+n3)×10

[0024] Among them, n1 is the number of neurons in the input layer, n2 is the number of neurons in the hidden layer, and n3 is the number of neurons in the output layer;

[0025] S33, decoding to obtain weights and thresholds, assigning the weights and thresholds to the BP neural network; using the training set to train the BP neural network, and using the test sample to verify the BP neural network; during the training process, the objective function is composed of the error matrix norm between the test value and the expected value of the test sample to obtain the minimum residual between the predicted value and the expected value;

[0026] S34, calculating fitness; wherein the fitness function is: Fitn V = ranking (obj), obj is the function output;

[0027] S35, select chromosomes with high fitness for replication;

[0028] S36, using single-point crossover to randomly generate new offspring individuals;

[0029] S37, randomly select mutant genes, and the selection operator uses random scan sampling;

[0030] S38, construct a new population;

[0031] S39, determine whether the minimum residual between the predicted value and the expected value meets the termination condition. If so, decode to obtain the optimal solution of the weight and threshold of the BP neural network, and output the trained BP neural network as the LNG tank rollover prediction model. Otherwise, go to step S33.

[0032] In a second aspect, the present invention discloses a LNG anti-rollover prediction device based on a GA-BP neural network algorithm, the device comprising:

[0033] The two-dimensional model building module is used to build a two-dimensional model of the LNG storage tank using Fluent software. The liquid layers in the two-dimensional model of the LNG storage tank have the same height and the liquid density in the layers is uniform; the two-dimensional model of the LNG storage tank is meshed and the initial conditions and boundary conditions are set;

[0034] The data set construction module is used to simulate the critical density difference corresponding to different tank volumes, diameters, initial density differences, and layer heights based on the two-dimensional model of the LNG tank, construct a tumbling process data set, and divide the tumbling process data set into a training set and a test set;

[0035] The prediction model training module is used to establish the LNG tank rollover prediction model based on the GA-BP neural network algorithm, and the training set is used to train the LNG tank rollover prediction model. During the training process, the minimum error between the test value and the expected value of the test sample is used as the objective function to adjust and optimize the parameters of the LNG tank rollover prediction model;

[0036] The prediction module is used to use the optimized and trained LNG tank rollover prediction model to perform rollover prediction, and calculate the critical density difference according to the input values ​​of the tank volume, diameter, and layer height to determine whether the LNG tank is about to roll over.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] First, the LNG anti-rollover prediction method and device based on the GA-BP neural network algorithm of the present invention simplifies the storage tank into a two-dimensional model, which reduces the computing resources and time required for simulation compared to the three-dimensional model. In addition, the complexity of the storage tank in the third direction, such as the vertical height and top structure of the storage tank, can be ignored, so as to focus on studying the flow and rollover phenomenon of the storage tank in the plane.

[0039] Second, the LNG anti-rollover prediction method and device based on the GA-BP neural network algorithm of the present invention uses a genetic algorithm to optimize the structure and parameters of the neural network, and through continuous evolution and selection, finds the optimal network structure and weights, so that the neural network can better approximate the objective function and improve the prediction accuracy and generalization ability of the model. The present invention combines the genetic algorithm with the neural network, optimizes the neural network through the genetic algorithm, and achieves fast convergence speed and obtains the global optimal effect.

[0040] Third, the LNG anti-rollover prediction method and device based on the GA-BP neural network algorithm of the present invention can understand the influence of different design parameters and operating conditions on the occurrence of rollover through in-depth research and simulation of the rollover phenomenon. The design and operation of the storage tank can be optimized to reduce the risk of rollover by adjusting the parameters such as the geometry of the storage tank, material selection, and filling method. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the relevant monitoring points set up for the simplified physical model of the storage tank;

[0042] Figure 2 is a flow chart of obtaining data through Fluent numerical simulation in the present invention;

[0043] Figure 3 It is a flow chart of the LNG anti-rollover prediction model based on the GA-BP neural network algorithm proposed in the present invention;

[0044] Figure 4 The diameter is 80m, 160,000m 3 Take the tank as an example, the relationship between the rollover coefficient and the initial density difference;

[0045] Figure 5 It is a schematic diagram of the accuracy of the prediction results of the GA-BP neural network algorithm model of the present invention, wherein (a) is the prediction accuracy of the training samples of the GA-BP neural network algorithm, and (b) is the prediction accuracy of the test samples of the GA-BP neural network algorithm. DETAILED DESCRIPTION

[0046] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0047] The present invention discloses a LNG anti-rollover prediction method based on a GA-BP neural network algorithm, the method comprising the following steps:

[0048] S1, Fluent software is used to construct a two-dimensional model of the LNG storage tank, in which the liquid layers have the same height and the liquid density within the layers is uniform; the two-dimensional model of the LNG storage tank is meshed, and initial conditions and boundary conditions are set;

[0049] S2, based on the simulation of the two-dimensional model of LNG storage tanks, obtain the critical density difference corresponding to different tank volumes, diameters, initial density differences, and layer heights, construct a tumbling process data set, and divide the tumbling process data set into a training set and a test set;

[0050] S3, based on the GA-BP neural network algorithm, the LNG tank rollover prediction model is established, and the training set is used to train the LNG tank rollover prediction model. During the training process, the minimum error between the test value and the expected value of the test sample is used as the objective function to adjust and optimize the parameters of the LNG tank rollover prediction model;

[0051] S4, using the optimized trained LNG tank rollover prediction model to perform rollover prediction, calculate the critical density difference according to the input values ​​of the tank volume, diameter, and layer height to determine whether the LNG tank is about to roll over.

[0052] Specifically, the prediction method includes the following steps:

[0053] 1. The present invention uses Fluent software to simulate the critical density difference of the working condition.

[0054] The present invention first intercepts two adjacent and equal-height layers of liquid in the LNG storage tank, and the density of the upper LNG layer is less than that of the lower LNG layer. The two liquid layers have the same height and the liquid density in the layers is uniform. Therefore, the LNG storage tank is simplified into a two-dimensional model structure as shown in FIG. Figure 1 , and set 9 monitoring points on the model (the monitoring point locations are as follows Figure 1 1 to 9 in the figure). Fluent software is used to construct the above LNG tank two-dimensional model, and then meshing is performed to set the initial conditions and boundary conditions. In the simulation process, only the tumbling process is considered, and the stratification process is ignored, so the effect of heat transfer on density change is ignored. Finally, the control parameters of the solver are set and the calculation is started. Figure 2 The figure is a flowchart of Fluent numerical simulation. Finally, the tumbling process data set obtained by Fluent simulation contains the conditions for tumbling and the key information of critical density difference, and this is established as the input and output indicator system of the neural network algorithm. In order to more accurately evaluate the performance of the model and improve the algorithm, the present invention divides the data set into a training set and a test set.

[0055] The critical density difference is related to the tank volume, diameter, layer height, and initial density difference. When the tank volume, diameter, and layer height remain unchanged, but the initial density difference changes, the tumbling coefficient and tumbling time corresponding to the tumbling will change. Then, the critical density difference of the tank is analyzed based on the tumbling coefficients corresponding to different initial density differences. If the tank volume, diameter, and layer height change, the tumbling coefficient and tumbling time corresponding to the tumbling will be different under the same initial density difference. In this embodiment, a tank volume of 160,000 m 3 , with a diameter of 80m; the simulation conditions are set to a layer height of 1m and an initial density difference of 1kg / m 3 The density of the upper layer of the tank is set to 430kg / m 3 , the lower layer density is 431kg / m 3 The Fluent multiphase flow model uses the mixture model, the turbulence equation uses the k-ε equation, and the boundary condition is set to the wall. The above conditions are brought into the Fluent software to simulate the initial density difference of 1kg / m 3 The tumbling time and tumbling coefficient are changed, and the initial density difference is changed to 2~9kg / m 3 (The difference is 1kg / m 3 ), and then get the corresponding data. Analyze the data at different initial density differences to find the critical density difference for safe storage in the tank. Figure 3 Indicates that the initial density difference is 0~3kg / m 3 When the density difference is greater than 3kg / m 3 When the tumbling coefficient increases sharply, it indicates that the tumbling degree of the two layers of LNG in the tank is severe. Therefore, the critical density difference for safe storage in a 160,000 cubic meter tank is 3 kg / m 3 In the storage tank, the stratification height of LNG is uncertain and the diameter will also change. Therefore, the stratification height when the LNG storage tank rolls will be changed. Other initial conditions remain unchanged, and the critical density difference at different stratification heights or different diameter lengths will be simulated and analyzed.

[0056] 2. The present invention uses genetic algorithm to optimize BP neural network to establish a rollover prediction model.

[0057] The present invention establishes a LNG tank rollover prediction model based on the GA-BP neural network algorithm, and uses a training set to train the model. During the training process, the parameters of the model are continuously adjusted to optimize its performance. Finally, a test set is used to test the final performance of the model. The present invention combines the mechanism of rollover and the factors causing rollover to establish a rollover prediction model based on the GA-BP neural network algorithm, such as Figure 4The flowchart of GA-BP neural network algorithm is shown in Figure 1. The input index system is the tank volume, diameter, and layer height, and the output index system is the critical density difference.

[0058] like Figure 4 As shown, first encode the original data and establish the initial population. The parameters optimized by the genetic algorithm refer to the initial weight and threshold of the BP neural network, determine the structure of the network, and obtain accurate data. The formula for the length of a single chromosome N is:

[0059] N=(n1×n2+n2×n3+n2+n3)×10

[0060] Among them, n1 is the number of neurons in the input layer, n2 is the number of neurons in the hidden layer, and n3 is the number of neurons in the output layer.

[0061] In the present invention, the weights and thresholds of the BP neural network are firstly encoded and an initial population is generated. After decoding, the weights and thresholds are assigned to the newly created BP neural network, and the network is trained. The training samples are used to train the neural network, and the test samples are used to verify the network performance and calculate the test error.

[0062] In the present invention, the objective function is composed of the error matrix norm between the test value and the expected value of the test sample to obtain the minimum residual between the predicted value and the expected value. Fitness function function: Fitn·V=ranking(obj), where obj is the function output.

[0063] The genetic algorithm calculates the fitness of the population, selects chromosomes with high fitness for replication, and uses random scanning sampling (SUS) as the selection operator. Subsequently, a single-point crossover operation is performed on the selected chromosomes to generate new offspring individuals. Next, by randomly selecting mutant genes, 1 becomes 0, or 0 becomes 1, the diversity of the population is further improved to form a new population.

[0064] After the new population is decoded, the weights and thresholds are assigned to the BP neural network again, the network is retrained and tested, and the test error is calculated. This process is iterated continuously until the termination condition is met (such as the error is less than the set threshold or the maximum number of iterations is reached). At this time, the decoding outputs the optimal BP network weights and thresholds, and the optimal solution for the required parameters is obtained, completing the genetic algorithm optimization process.

[0065] The network model established by the present invention adopts the newff function, which can randomly divide the training data into two groups after disrupting the order, of which 85% are training samples and 15% are verification samples, which are independent of the test samples and have no overlap. The training samples are used to determine the nonlinear relationship between the network input and output, and the verification samples are used to verify the generalization ability of the network. With the increase of the number of training times, the training is stopped when the error of the verification sample does not decrease. In essence, it is equivalent to an early termination strategy, which can prevent the network from overfitting.

[0066] Since different population sizes, different evolutionary generations, and different hidden layers will all have an impact on the training results, we first use empirical formulas to determine the number of hidden layers to be 9 and the number of evolutionary generations to be 400. For different models, these values ​​may need to be adjusted, so we first conduct preliminary training: when the number of evolutionary generations is 400, the number of hidden layers is 9, and the population size is 100-700, the best result is obtained when the population size is 400, the number of evolutionary generations is 400, and the number of hidden layers is 2-16, the best result is obtained when the number of hidden layers is 9; when the population size is 400, the number of hidden layers is 9, and the number of evolutionary generations is 100-700, the best result is obtained when the number of evolutionary generations is 400. Therefore, it is concluded that the best training effect is obtained when the population size is 400, the number of hidden layers is 9, and the number of evolutionary generations is 400.

[0067] After preliminary investigation, the training range becomes: population size 300-500; hidden layer number 5, 9, 13; evolutionary generations 300-500, cross-combination, the following are 83 training results after cross-combination. Table 1 shows the training results under different evolutionary generations and different hidden layers when the population is 300; Table 2 shows the training results under different evolutionary generations and different hidden layers when the population is 400; Table 4 shows the training results under different evolutionary generations and different hidden layers when the population is 500.

[0068] Table 1

[0069]

[0070] Table 2

[0071]

[0072] Table 3

[0073]

[0074]

[0075] When the evolutionary generations are 300, 400, and 500, and the number of hidden layers is 5, the simulation error of the test sample increases first and then decreases, but the decrease in the simulation error of the test sample is small when the evolutionary generation changes from 400 to 500, indicating that when the population size is 500, the increase in the evolutionary generation will no longer affect the simulation error of the test sample, and overfitting occurs. Through the above analysis, it can be concluded that the training results are best when the population size is 400, the evolutionary generation is 400, and the number of hidden layers is 9, and the simulation error of the test sample is the smallest.

[0076] Therefore, the genetic neural network parameters are determined as follows:

[0077] Determination of some parameters of the genetic algorithm: the population size is 400, the evolutionary generation is 400, and other parameters are obtained based on experience: the selection probability is 0.95, the crossover probability is 0.7, the mutation probability is 0.01, and the chromosome length is calculated to be 75.

[0078] Determination of some parameters of the neural network: the number of training samples is 60, the number of test samples is 23, the learning rate is 0.1, the number of training times is 1000, the training target is 0.01, the number of neurons in the input layer is 4, the number of neurons in the hidden layer is 9, and the number of neurons in the output layer is 1.

[0079] The trained GA-BP neural network algorithm is used to predict rollover, and the critical density difference is calculated according to the value of the input index system to determine whether the LNG storage tank is about to roll over. Figure 5 The prediction accuracy of the training samples and test samples of the GA-BP neural network algorithm. From the figure, we can see the prediction results of the critical density difference of the training set and the test set. Overall, the error of the training results is small, which can be used to predict the critical density difference of LNG storage tanks.

[0080] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0084] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0085] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A LNG anti-rollover prediction method based on GA-BP neural network algorithm, characterized in that: The method comprises the following steps: S1, Fluent software is used to construct a two-dimensional model of the LNG storage tank, in which the liquid layers have the same height and the liquid density within the layers is uniform; the two-dimensional model of the LNG storage tank is meshed, and initial conditions and boundary conditions are set; S2, based on the simulation of the two-dimensional model of LNG storage tanks, obtain the critical density difference corresponding to different tank volumes, diameters, initial density differences, and layer heights, construct a tumbling process data set, and divide the tumbling process data set into a training set and a test set; S3, based on the GA-BP neural network algorithm, the LNG tank rollover prediction model is established, and the training set is used to train the LNG tank rollover prediction model. During the training process, the minimum error between the test value and the expected value of the test sample is used as the objective function to adjust and optimize the parameters of the LNG tank rollover prediction model; S4, using the optimized trained LNG tank rollover prediction model to perform rollover prediction, calculate the critical density difference according to the input tank volume, diameter, and layer height values ​​to determine whether the LNG tank is about to roll over.

2. The LNG anti-rollover prediction method based on GA-BP neural network algorithm according to claim 1 is characterized in that: Step S1 further comprises: Intercept two adjacent and equal-height stratified liquids in the LNG storage tank, where the density of the upper LNG layer is less than that of the lower LNG layer, the two liquid strata have the same height, and the density of the liquids in the strata is uniform. Simplify the LNG storage tank into a two-dimensional solid model, and set multiple monitoring points on the two-dimensional solid model. Based on the simplified two-dimensional solid model, the Fluent software was used to construct a two-dimensional model of the LNG storage tank based on the mixture model and perform meshing; Set the initial conditions and boundary conditions; the initial conditions include the tank volume, diameter, layer height and initial density difference; the boundary condition is set to the wall, and the turbulence equation uses the k-ε equation.

3. The LNG anti-rollover prediction method based on GA-BP neural network algorithm according to claim 1 is characterized in that: Step S2 further comprises: S21, set the tank volume, diameter and layer height; S22, setting multiple groups of initial density differences to simulate the tumbling time and tumbling coefficient of different initial density differences; S23, analyzing the variation range of the rollover coefficient corresponding to different initial density differences, and setting the initial density difference when the variation range of the rollover coefficient exceeds a preset variation threshold as the critical density difference for safe storage of the storage tank; S24, changing the tank volume, diameter and layer height, repeating steps S22 to S23, until critical density differences corresponding to multiple groups of tank volumes, diameters and layer heights are obtained, and a tumbling process data set is constructed.

4. The LNG anti-rollover prediction method based on GA-BP neural network algorithm according to claim 1 is characterized in that: Step S3 further comprises: S31, determine the topological structure of the GA-BP neural network; S32, encode the original data and establish the initial population; the length N of a single chromosome is: N=(n1×n2+n2×n3+n2+n3)×10 Among them, n1 is the number of neurons in the input layer, n2 is the number of neurons in the hidden layer, and n3 is the number of neurons in the output layer; S33, decoding to obtain weights and thresholds, assigning the weights and thresholds to the BP neural network; using the training set to train the BP neural network, and using the test sample to verify the BP neural network; during the training process, the objective function is composed of the error matrix norm between the test value and the expected value of the test sample to obtain the minimum residual between the predicted value and the expected value; S34, calculating fitness; wherein the fitness function is: FitnV=ranking(obj), obj is the function output; S35, select chromosomes with high fitness for replication; S36, using single-point crossover to randomly generate new offspring individuals; S37, randomly select mutant genes, and the selection operator uses random scan sampling; S38, construct a new population; S39, determine whether the minimum residual between the predicted value and the expected value meets the termination condition. If so, decode to obtain the optimal solution of the weight and threshold of the BP neural network, and output the trained BP neural network as the LNG tank rollover prediction model. Otherwise, go to step S33.

5. An LNG anti-rollover prediction device based on GA-BP neural network algorithm, characterized in that: The device comprises: The two-dimensional model building module is used to build a two-dimensional model of the LNG storage tank using Fluent software. The liquid layers in the two-dimensional model of the LNG storage tank have the same height and the liquid density in the layers is uniform; the two-dimensional model of the LNG storage tank is meshed and the initial conditions and boundary conditions are set; The data set construction module is used to simulate the critical density difference corresponding to different tank volumes, diameters, initial density differences, and layer heights based on the two-dimensional model of the LNG tank, construct a tumbling process data set, and divide the tumbling process data set into a training set and a test set; The prediction model training module is used to establish the LNG tank rollover prediction model based on the GA-BP neural network algorithm, and the training set is used to train the LNG tank rollover prediction model. During the training process, the minimum error between the test value and the expected value of the test sample is used as the objective function to adjust and optimize the parameters of the LNG tank rollover prediction model; The prediction module is used to use the optimized and trained LNG tank rollover prediction model to perform rollover prediction, and calculate the critical density difference according to the input values ​​of the tank volume, diameter, and layer height to determine whether the LNG tank is about to roll over.

Images